EdgeAI YOLOv8 Desktop App

A C++ desktop application for YOLOv8 object detection using Qt, OpenCV, and ONNX Runtime. The app supports webcam inference and single-image inference, and prints runtime metrics for comparison with the Python implementation.

Features

• Real-time webcam object detection
• Static image detection through a file picker
• YOLOv8 ONNX Runtime inference on CPU
• COCO class labels with confidence scores
• Camera refresh and camera selection controls
• Console performance metrics for FPS, inference latency, CPU time, and memory use

Project Structure

.
├── CMakeLists.txt
├── models/
│   └── yolov8n.onnx
├── onnxruntime-linux-x64-1.20.1/
├── screenshots/
│   ├── home_page.png
│   └── live_inferencing.png
├── src/
│   ├── detector/
│   ├── gui/
│   └── main.cpp
└── yaml/
    └── coco.names

Requirements

• Linux x86_64
• CMake 3.16 or newer
• C++17 compiler
• Qt5 Widgets
• OpenCV
• ONNX Runtime 1.20.1

On Ubuntu/Debian:

sudo apt update
sudo apt install -y build-essential cmake qtbase5-dev libopencv-dev

Build

cmake -S . -B build
cmake --build build -j$(nproc)

Run

./build/edgeai-app

The app loads:

• Model: models/yolov8n.onnx
• Classes: yaml/coco.names

Performance Logs

The app prints comparison-friendly metrics:

[C++] YOLO model load | model=... | load_ms=... | ready=true
[C++] Webcam performance | frames=30 | fps=... | avg_frame_ms=... | avg_inference_ms=... | detections=... | cpu_ms=... | max_rss_mb=...
[C++] Image performance | file=... | total_ms=... | inference_ms=... | detections=... | process_cpu_s=... | max_rss_mb=...

Use these values alongside the Python app logs for runtime comparison.

Notes

• OpenCV may print warnings while probing unavailable camera indexes.
• build/ is generated locally and ignored by git.
• The current CMake file links against onnxruntime-linux-x64-1.20.1/lib/libonnxruntime.so.